Research On Visual Map Construction And High Precision Positioning For Intelligent Connected Vehicles

Intelligent connected vehicles refer to the organic combination of the Intrnet of Vehicles and intelligent vehicles,where vehicles and external nodes such as maps and other vehicles achieve environmental awareness,information sharing,intelligent decision-making,and collaborative control to achieve safe,orderly,and efficient driving,ultimately replacing human operation as a new generation of vehicles.And high-precision positioning is one of the key steps to achieve environmental awareness.For the high-precision positioning of intelligent connected vehicles,this article adopts the research method of "first collecting map featuresthen conducting high-precision positioning".Based on the visual map model,monocular vision is combined with integrated inertial navigation systems to study the high-precision positioning method of intelligent connected vehicles.The research results of this article can effectively meet the low-cost and high-precision positioning needs of intelligent connected vehicles,and are of great significance for promoting the development of intelligent connected vehicles.The main research work of this article is as follows:Firstly,a calibration method for vehicle mounted cameras based on a checkerboard calibration board was proposed.It establishes a correspondence between the world coordinate system and the pixel coordinate system,constructs a pixel coordinate system for the photos collected by the camera,and takes any point to calculate the position of the corresponding point in the world coordinate system.The position error range can be as low as 5cm,thus achieving high-precision positioning of intelligent connected vehicles.Secondly,a visual map model consisting of a series of nodes is established using a monocular camera and a dual antenna differential GPS sensor.Each node contains three levels:GPS information,image features,and 3D information.Among them,image features mainly refer to local features in the upper half(distant view)of the forward looking image;The threedimensional information data consists of the lower half of the forward view image(close range)and the relative position relationship between the vehicle and the road markings obtained from field measurements.The existence of three levels ensures the uniqueness of each node and improves positioning accuracy.Finally,a multi-level positioning method for intelligent networked vehicles based on inertial navigation and image fusion was proposed.Based on the visual map model,the monocular vision is combined with the integrated inertial navigation system to achieve multilevel positioning using the method of "GPS initial positioning-long-range positioning-close range positioning".At the initial GPS positioning level,compare the GPS information collected by inertial navigation with the GPS information in the visual map to obtain the nodes within the initial positioning range;At the perspective level localization level,extract the perspective features of a monocular image and match them with the image features of the initial localization node to obtain the nodes for the perspective level localization;At the close range level positioning level,a camera is used to calibrate and measure the relative position relationship between road markings and vehicles,and then the position of the intelligent connected vehicle relative to this node is matched with the three-dimensional information data in the distant level positioning node.Finally,Kalman filtering is used to optimize the positioning results.The experimental results show that compared with using Kalman filter and not using Kalman filter,the vertical positioning accuracy has been improved by 14.2%,and the horizontal positioning accuracy has been improved by 34.1%.